@absolutejs/absolute 0.19.0-beta.618 → 0.19.0-beta.619

package/dist/ai/index.js

@@ -4237,6 +4237,25 @@ var buildContextLabel2 = (metadata) => {
   if (!metadata) {
     return;
   }
+  const pdfTextKind = getContextString2(metadata.pdfTextKind);
+  const officeBlockKind = getContextString2(metadata.officeBlockKind);
+  const sectionPath = Array.isArray(metadata.sectionPath) ? metadata.sectionPath.map((value) => getContextString2(value)).filter((value) => typeof value === "string") : [];
+  const sectionTitle = getContextString2(metadata.sectionTitle) ?? sectionPath.at(-1);
+  if (pdfTextKind === "table_like" && sectionTitle) {
+    return `PDF table block ${sectionTitle}`;
+  }
+  if (pdfTextKind === "paragraph" && sectionTitle) {
+    return `PDF text block ${sectionTitle}`;
+  }
+  if (officeBlockKind === "table" && sectionTitle) {
+    return `Office table block ${sectionTitle}`;
+  }
+  if (officeBlockKind === "list" && sectionTitle) {
+    return `Office list block ${sectionTitle}`;
+  }
+  if (officeBlockKind === "paragraph" && sectionTitle) {
+    return `Office paragraph block ${sectionTitle}`;
+  }
   const emailKind = getContextString2(metadata.emailKind);
   if (emailKind === "attachment") {
     return "Attachment evidence";
@@ -4273,8 +4292,6 @@ var buildContextLabel2 = (metadata) => {
   if (speaker) {
     return `Speaker ${speaker}`;
   }
-  const sectionPath = Array.isArray(metadata.sectionPath) ? metadata.sectionPath.map((value) => getContextString2(value)).filter((value) => typeof value === "string") : [];
-  const sectionTitle = getContextString2(metadata.sectionTitle) ?? sectionPath.at(-1);
   if (sectionTitle) {
     return `Section ${sectionTitle}`;
   }
@@ -4284,11 +4301,21 @@ var buildLocatorLabel2 = (metadata, source, title) => {
   if (!metadata) {
     return;
   }
+  const pdfTextKind = getContextString2(metadata.pdfTextKind);
+  const officeBlockKind = getContextString2(metadata.officeBlockKind);
+  const pdfBlockNumber = getContextNumber2(metadata.pdfBlockNumber);
+  const officeBlockNumber = getContextNumber2(metadata.officeBlockNumber);
   const page = getContextNumber2(metadata.page) ?? getContextNumber2(metadata.pageNumber) ?? (typeof metadata.pageIndex === "number" ? metadata.pageIndex + 1 : undefined);
   const region = getContextNumber2(metadata.regionNumber) ?? (typeof metadata.regionIndex === "number" ? metadata.regionIndex + 1 : undefined);
   if (page && region) {
     return `Page ${page} \xB7 Region ${region}`;
   }
+  if (page && pdfBlockNumber && pdfTextKind === "table_like") {
+    return `Page ${page} \xB7 Table Block ${pdfBlockNumber}`;
+  }
+  if (page && pdfBlockNumber) {
+    return `Page ${page} \xB7 Text Block ${pdfBlockNumber}`;
+  }
   if (page) {
     return `Page ${page}`;
   }
@@ -4317,6 +4344,15 @@ var buildLocatorLabel2 = (metadata, source, title) => {
   if (mediaStart) {
     return `Timestamp ${mediaStart}`;
   }
+  if (officeBlockNumber && officeBlockKind === "table") {
+    return `Office table block ${officeBlockNumber}`;
+  }
+  if (officeBlockNumber && officeBlockKind === "list") {
+    return `Office list block ${officeBlockNumber}`;
+  }
+  if (officeBlockNumber && officeBlockKind === "paragraph") {
+    return `Office paragraph block ${officeBlockNumber}`;
+  }
   const sectionPath = Array.isArray(metadata.sectionPath) ? metadata.sectionPath.map((value) => getContextString2(value)).filter((value) => typeof value === "string") : [];
   if (sectionPath.length > 0) {
     return `Section ${sectionPath.join(" > ")}`;
@@ -4334,12 +4370,16 @@ var buildProvenanceLabel2 = (metadata) => {
   const mediaKind = getContextString2(metadata.mediaKind);
   const transcriptSource = getContextString2(metadata.transcriptSource);
   const pdfTextMode = getContextString2(metadata.pdfTextMode);
+  const pdfTextKind = getContextString2(metadata.pdfTextKind);
+  const officeBlockKind = getContextString2(metadata.officeBlockKind);
   const ocrEngine = getContextString2(metadata.ocrEngine);
   const extractorRegistryMatch = getContextString2(metadata.extractorRegistryMatch);
   const chunkingProfile = getContextString2(metadata.chunkingProfile);
   const ocrConfidence = getContextNumber2(metadata.ocrRegionConfidence) ?? getContextNumber2(metadata.ocrConfidence);
   const labels = [
     pdfTextMode ? `PDF ${pdfTextMode}` : "",
+    pdfTextKind === "table_like" ? "PDF table block" : pdfTextKind === "paragraph" ? "PDF text block" : "",
+    officeBlockKind ? `Office ${officeBlockKind}` : "",
     ocrEngine ? `OCR ${ocrEngine}` : "",
     extractorRegistryMatch ? `Extractor ${extractorRegistryMatch}` : "",
     chunkingProfile ? `Chunking ${chunkingProfile}` : "",
@@ -4375,7 +4415,7 @@ var buildRAGChunkStructure = (metadata) => {
     return;
   }
   const sectionPath = Array.isArray(metadata.sectionPath) ? metadata.sectionPath.filter((value) => typeof value === "string" && value.trim().length > 0) : undefined;
-  const sectionKind = metadata.sectionKind === "markdown_heading" || metadata.sectionKind === "html_heading" || metadata.sectionKind === "office_heading" || metadata.sectionKind === "spreadsheet_rows" || metadata.sectionKind === "presentation_slide" ? metadata.sectionKind : undefined;
+  const sectionKind = metadata.sectionKind === "markdown_heading" || metadata.sectionKind === "html_heading" || metadata.sectionKind === "office_heading" || metadata.sectionKind === "office_block" || metadata.sectionKind === "pdf_block" || metadata.sectionKind === "spreadsheet_rows" || metadata.sectionKind === "presentation_slide" ? metadata.sectionKind : undefined;
   const section = {
     depth: getContextNumber2(metadata.sectionDepth),
     kind: sectionKind,
@@ -4695,7 +4735,7 @@ var buildRAGSourceSummaries = (sources) => {
   const citationReferenceMap = buildRAGCitationReferenceMap(citations);
   return sourceGroups.map((group) => {
     const groupCitations = citations.filter((citation) => group.chunks.some((chunk) => chunk.chunkId === citation.chunkId));
-    const leadChunk = group.chunks.slice().sort((left, right) => right.score - left.score)[0];
+    const leadChunk = getPreferredSourceLeadChunk(group.chunks);
     const excerpts = leadChunk ? buildRAGChunkExcerpts(group.chunks, leadChunk.chunkId) : undefined;
     const structure = leadChunk?.structure ?? buildRAGChunkStructure(leadChunk?.metadata);
     const excerptSelection = buildRAGExcerptSelection(excerpts, structure);
@@ -4723,13 +4763,45 @@ var getSectionPathFromSource = (source) => {
   const path = source.structure?.section?.path ?? (Array.isArray(source.metadata?.sectionPath) ? source.metadata.sectionPath.map((value) => getContextString2(value)).filter((value) => typeof value === "string") : []);
   return path.length > 0 ? path : undefined;
 };
+var isBlockAwareContextLabel = (value) => typeof value === "string" && (value.startsWith("PDF ") || value.startsWith("Office "));
+var getStructuredSectionScoreWeight = (metadata) => {
+  if (!metadata) {
+    return 1;
+  }
+  const pdfTextKind = getContextString2(metadata.pdfTextKind);
+  const officeBlockKind = getContextString2(metadata.officeBlockKind);
+  const sectionKind = getContextString2(metadata.sectionKind);
+  if (pdfTextKind === "table_like") {
+    return 1.28;
+  }
+  if (officeBlockKind === "table" || officeBlockKind === "list") {
+    return 1.24;
+  }
+  if (sectionKind === "pdf_block" || sectionKind === "office_block" || officeBlockKind === "paragraph" || pdfTextKind === "paragraph") {
+    return 1.12;
+  }
+  return 1;
+};
+var getStructuredSourceLeadScore = (source) => source.score * getStructuredSectionScoreWeight(source.metadata);
+var getPreferredSourceLeadChunk = (chunks) => chunks.slice().sort((left, right) => {
+  const leftWeightedScore = getStructuredSourceLeadScore(left);
+  const rightWeightedScore = getStructuredSourceLeadScore(right);
+  if (rightWeightedScore !== leftWeightedScore) {
+    return rightWeightedScore - leftWeightedScore;
+  }
+  if (right.score !== left.score) {
+    return right.score - left.score;
+  }
+  return left.chunkId.localeCompare(right.chunkId);
+})[0];
 var buildRAGSectionRetrievalDiagnostics = (sources, trace) => {
-  const totalScore = sources.reduce((sum, source) => sum + source.score, 0);
+  const totalScore = sources.reduce((sum, source) => sum + source.score * getStructuredSectionScoreWeight(source.metadata), 0);
   if (sources.length === 0 || totalScore <= 0) {
     return [];
   }
   const sections = new Map;
   for (const source of sources) {
+    const structuredScore = source.score * getStructuredSectionScoreWeight(source.metadata);
     const path = getSectionPathFromSource(source);
     if (!path) {
       continue;
@@ -4761,7 +4833,7 @@ var buildRAGSectionRetrievalDiagnostics = (sources, trace) => {
         sourceSet: new Set(source.source ? [source.source] : []),
         topChunkId: source.chunkId,
         topSource: source.source,
-        totalScore: source.score,
+        totalScore: structuredScore,
         transformedHits,
         variantHits,
         vectorHits
@@ -4769,7 +4841,7 @@ var buildRAGSectionRetrievalDiagnostics = (sources, trace) => {
       continue;
     }
     existing.count += 1;
-    existing.totalScore += source.score;
+    existing.totalScore += structuredScore;
     if (source.source) {
       existing.sourceSet.add(source.source);
     }
@@ -4797,6 +4869,8 @@ var buildRAGSectionRetrievalDiagnostics = (sources, trace) => {
     const parentTotal = siblingPool.reduce((sum, entry) => sum + entry.totalScore, 0);
     const scoreShare = section.totalScore / totalScore;
     const parentShare = parentTotal > 0 ? section.totalScore / parentTotal : undefined;
+    const topChunk = sources.find((source) => source.chunkId === section.topChunkId);
+    const topContextLabel = topChunk?.labels?.contextLabel ?? buildContextLabel2(topChunk?.metadata);
     const parentDistribution = parentTotal > 0 ? siblingPool.map((entry) => ({
       count: entry.count,
       isActive: entry.key === section.key,
@@ -4922,6 +4996,7 @@ var buildRAGSectionRetrievalDiagnostics = (sources, trace) => {
       reasons.push("concentrated_evidence");
     }
     const summaryParts = [
+      isBlockAwareContextLabel(topContextLabel) ? topContextLabel : "",
       `${section.count} hit${section.count === 1 ? "" : "s"}`,
       `${(scoreShare * 100).toFixed(0)}% score share`,
       `vector ${section.vectorHits} \xB7 lexical ${section.lexicalHits} \xB7 hybrid ${section.hybridHits}`,
@@ -5133,22 +5208,21 @@ var updateSourceGroup = (groups, source) => {
     groups.set(key, buildSourceGroup(source, key));
     return;
   }
-  if (source.score > existing.bestScore) {
-    existing.bestScore = source.score;
-    existing.label = buildSourceLabel2(source);
-    existing.labels = source.labels ?? buildRAGSourceLabels({
-      metadata: source.metadata,
-      source: source.source,
-      title: source.title
-    });
-    existing.structure = source.structure ?? buildRAGChunkStructure(source.metadata);
-    existing.source = source.source;
-    existing.title = source.title;
-  } else {
-    existing.bestScore = Math.max(existing.bestScore, source.score);
-  }
+  existing.bestScore = Math.max(existing.bestScore, source.score);
   existing.count += 1;
   existing.chunks.push(source);
+  const leadChunk = getPreferredSourceLeadChunk(existing.chunks);
+  if (leadChunk) {
+    existing.label = buildSourceLabel2(leadChunk);
+    existing.labels = leadChunk.labels ?? buildRAGSourceLabels({
+      metadata: leadChunk.metadata,
+      source: leadChunk.source,
+      title: leadChunk.title
+    });
+    existing.structure = leadChunk.structure ?? buildRAGChunkStructure(leadChunk.metadata);
+    existing.source = leadChunk.source;
+    existing.title = leadChunk.title;
+  }
 };
 var getLatestAssistantMessage = (messages) => {
   for (let index = messages.length - 1;index >= 0; index -= 1) {
@@ -8485,6 +8559,55 @@ var scoreLoosePhraseMatch2 = (query, text) => {
   }
   return 0;
 };
+var queryHasAnyToken = (queryTokens, candidates) => candidates.some((candidate) => queryTokens.includes(candidate));
+var scoreStructuredEvidenceMatch = (queryTokens, result) => {
+  const metadata = result.metadata ?? {};
+  const pdfTextKind = typeof metadata.pdfTextKind === "string" ? metadata.pdfTextKind : undefined;
+  const officeBlockKind = typeof metadata.officeBlockKind === "string" ? metadata.officeBlockKind : undefined;
+  const hasBlockMetadata = typeof metadata.pdfBlockNumber === "number" || typeof metadata.officeBlockNumber === "number";
+  let score = 0;
+  if (hasBlockMetadata) {
+    score += 0.12;
+  }
+  if (pdfTextKind === "table_like" && queryHasAnyToken(queryTokens, [
+    "table",
+    "row",
+    "rows",
+    "column",
+    "columns",
+    "spreadsheet",
+    "sheet",
+    "workbook"
+  ])) {
+    score += 0.65;
+  }
+  if (officeBlockKind === "table" && queryHasAnyToken(queryTokens, [
+    "table",
+    "row",
+    "rows",
+    "column",
+    "columns",
+    "matrix",
+    "grid"
+  ])) {
+    score += 0.55;
+  }
+  if (officeBlockKind === "list" && queryHasAnyToken(queryTokens, [
+    "list",
+    "checklist",
+    "bullet",
+    "bullets",
+    "step",
+    "steps",
+    "task",
+    "tasks",
+    "item",
+    "items"
+  ])) {
+    score += 0.55;
+  }
+  return score;
+};
 var scoreHeuristicMatch = ({
   query,
   queryTokens,
@@ -8501,7 +8624,8 @@ var scoreHeuristicMatch = ({
   const exactPhraseBoost = Math.max(normalizeText([result.title, result.source, result.chunkText, ...metadataValues].filter(Boolean).join(" ")).includes(queryTokens.join(" ")) ? 1 : 0, scoreLoosePhraseMatch2(query, [result.title, result.source, result.chunkText, ...metadataValues].filter(Boolean).join(" ")));
   const sourcePathBoost = typeof result.source === "string" && queryTokens.some((token) => result.source?.toLowerCase().includes(token)) ? 0.5 : 0;
   const metadataBoost = metadataValues.length > 0 ? queryTokens.filter((token) => metadataValues.some((value) => value.toLowerCase().includes(token))).length / queryTokens.length : 0;
-  return result.score + overlapBoost + exactPhraseBoost + sourcePathBoost + metadataBoost;
+  const structuredEvidenceBoost = scoreStructuredEvidenceMatch(queryTokens, result);
+  return result.score + overlapBoost + exactPhraseBoost + sourcePathBoost + metadataBoost + structuredEvidenceBoost;
 };
 var normalizeText = (value) => tokenize3(value).join(" ");
 var applyRAGReranking = async ({
@@ -8736,32 +8860,59 @@ var stripHtmlTags = (value) => {
 `).replace(/<li\b[^>]*>/gi, "- ").replace(/<[^>]+>/g, " ");
   return decodeHtmlEntities(withoutTags);
 };
+var stripHtmlNoiseBlocks = (value) => value.replace(/<!--[\s\S]*?-->/g, " ").replace(/<(script|style|template|noscript)\b[^>]*>[\s\S]*?<\/\1>/gi, " ").replace(/<([a-z0-9:_-]+)\b[^>]*\b(hidden|aria-hidden=(['"])true\3)[^>]*>[\s\S]*?<\/\1>/gi, " ").replace(/<(nav|footer|header|aside|form|dialog)\b[^>]*>[\s\S]*?<\/\1>/gi, " ").replace(/<([a-z0-9:_-]+)\b[^>]*\b(?:id|class)=(['"])[^'"]*(nav|menu|footer|header|sidebar|promo|banner|cookie|breadcrumb|share|social|subscribe|newsletter|modal)[^'"]*\2[^>]*>[\s\S]*?<\/\1>/gi, " ");
+var collectHtmlContentCandidates = (value) => {
+  const patterns = [
+    {
+      contentGroup: 1,
+      pattern: /<main\b[^>]*>([\s\S]*?)<\/main>/gi
+    },
+    {
+      contentGroup: 1,
+      pattern: /<article\b[^>]*>([\s\S]*?)<\/article>/gi
+    },
+    {
+      contentGroup: 3,
+      pattern: /<([a-z0-9:_-]+)\b[^>]*\brole=(['"])main\2[^>]*>([\s\S]*?)<\/\1>/gi
+    },
+    {
+      contentGroup: 4,
+      pattern: /<([a-z0-9:_-]+)\b[^>]*\b(?:id|class)=(['"])[^'"]*(content|article|main|post|body)[^'"]*\2[^>]*>([\s\S]*?)<\/\1>/gi
+    }
+  ];
+  const candidates = [];
+  for (const entry of patterns) {
+    for (const match of value.matchAll(entry.pattern)) {
+      const rawCandidate = match[entry.contentGroup];
+      const candidate = typeof rawCandidate === "string" ? rawCandidate : "";
+      if (candidate.trim()) {
+        candidates.push(candidate.trim());
+      }
+    }
+  }
+  return candidates;
+};
 var extractMainHtmlContent = (value) => {
   const trimmed = value.trim();
   if (!/<html\b|<body\b|<main\b|<article\b/i.test(trimmed)) {
     return value;
   }
-  const boilerplateStripped = trimmed.replace(/<script\b[^>]*>[\s\S]*?<\/script>/gi, " ").replace(/<style\b[^>]*>[\s\S]*?<\/style>/gi, " ").replace(/<(nav|footer|header|aside|form)\b[^>]*>[\s\S]*?<\/\1>/gi, " ");
-  const mainMatch = boilerplateStripped.match(/<main\b[^>]*>([\s\S]*?)<\/main>/i);
-  if (mainMatch?.[1]) {
-    return mainMatch[1];
-  }
-  const articleMatches = [
-    ...boilerplateStripped.matchAll(/<article\b[^>]*>([\s\S]*?)<\/article>/gi)
-  ].map((match) => match[1]?.trim()).filter(Boolean);
-  if (articleMatches.length > 0) {
-    return articleMatches.join(`
-`);
-  }
-  const roleMainMatch = boilerplateStripped.match(/<([a-z0-9:_-]+)\b[^>]*\brole=(['"])main\2[^>]*>([\s\S]*?)<\/\1>/i);
-  if (roleMainMatch?.[3]) {
-    return roleMainMatch[3];
+  const stripped = stripHtmlNoiseBlocks(trimmed);
+  const candidates = collectHtmlContentCandidates(stripped);
+  if (candidates.length > 0) {
+    const bestCandidate = candidates.map((candidate) => ({
+      candidate,
+      score: stripHtmlTags(candidate).replace(/\s+/g, " ").trim().length
+    })).sort((left, right) => right.score - left.score)[0]?.candidate;
+    if (bestCandidate) {
+      return bestCandidate;
+    }
   }
-  const bodyMatch = boilerplateStripped.match(/<body\b[^>]*>([\s\S]*?)<\/body>/i);
+  const bodyMatch = stripped.match(/<body\b[^>]*>([\s\S]*?)<\/body>/i);
   if (bodyMatch?.[1]) {
     return bodyMatch[1];
   }
-  return boilerplateStripped;
+  return stripped;
 };
 var stripHtml = (value) => {
   const focused = extractMainHtmlContent(value);
@@ -8779,6 +8930,93 @@ var stripMarkdown = (value) => {
 `);
   return normalizeWhitespace(stripped);
 };
+var pdfNativeStructureUnits = (metadata) => {
+  const blocks = Array.isArray(metadata?.pdfTextBlocks) ? metadata.pdfTextBlocks : [];
+  const units = [];
+  for (const block of blocks) {
+    if (!block || typeof block !== "object") {
+      continue;
+    }
+    const text = typeof block.text === "string" ? normalizeWhitespace(block.text) : "";
+    if (!text) {
+      continue;
+    }
+    const pageNumber = typeof block.pageNumber === "number" && Number.isFinite(block.pageNumber) ? block.pageNumber : undefined;
+    const pdfBlockNumber = typeof block.blockNumber === "number" && Number.isFinite(block.blockNumber) ? block.blockNumber : undefined;
+    const pdfTextKind = block.textKind === "table_like" ? "table_like" : "paragraph";
+    const sectionTitle = pageNumber ? pdfTextKind === "table_like" ? `Page ${pageNumber} Table Block` : `Page ${pageNumber} Text Block` : pdfTextKind === "table_like" ? "Table Block" : "Text Block";
+    units.push({
+      pageNumber,
+      pdfBlockNumber,
+      pdfTextKind,
+      preferredChunkUnits: pdfTextKind === "table_like" ? text.split(`
+`).filter(Boolean) : undefined,
+      sectionDepth: 1,
+      sectionKind: "pdf_block",
+      sectionPath: [sectionTitle],
+      sectionTitle,
+      text
+    });
+  }
+  return units;
+};
+var officeNativeStructureUnits = (metadata) => {
+  const blocks = Array.isArray(metadata?.officeBlocks) ? metadata.officeBlocks : [];
+  const units = [];
+  const headingStack = [];
+  const decorateOfficeSectionText = (text, sectionTitle) => {
+    if (!sectionTitle || text.includes(sectionTitle)) {
+      return text;
+    }
+    return normalizeWhitespace(`${sectionTitle}
+${text}`);
+  };
+  for (const [index, block] of blocks.entries()) {
+    if (!block || typeof block !== "object") {
+      continue;
+    }
+    const text = typeof block.text === "string" ? normalizeWhitespace(block.text) : "";
+    if (!text) {
+      continue;
+    }
+    const officeBlockNumber = typeof block.blockNumber === "number" && Number.isFinite(block.blockNumber) ? block.blockNumber : undefined;
+    const officeBlockKind = block.blockKind === "title" || block.blockKind === "heading" || block.blockKind === "list" || block.blockKind === "table" ? block.blockKind : "paragraph";
+    const headingLevel = typeof block.headingLevel === "number" && Number.isFinite(block.headingLevel) ? block.headingLevel : undefined;
+    if (officeBlockKind === "title" || officeBlockKind === "heading") {
+      const level = officeBlockKind === "title" ? 1 : headingLevel ?? 1;
+      headingStack[level - 1] = text;
+      headingStack.length = level;
+      const nextBlock = blocks[index + 1];
+      const nextKind = nextBlock && typeof nextBlock === "object" ? nextBlock.blockKind : undefined;
+      if (nextKind === "title" || nextKind === "heading" || nextKind === "list" || nextKind === "table" || !nextBlock) {
+        units.push({
+          officeBlockKind,
+          officeBlockNumber,
+          sectionDepth: headingStack.length,
+          sectionKind: "office_heading",
+          sectionPath: [...headingStack],
+          sectionTitle: text,
+          text
+        });
+      }
+      continue;
+    }
+    const sectionPath = headingStack.length > 0 ? [...headingStack] : undefined;
+    const sectionTitle = sectionPath?.at(-1);
+    units.push({
+      officeBlockKind,
+      officeBlockNumber,
+      preferredChunkUnits: officeBlockKind === "table" ? text.split(`
+`).filter(Boolean) : undefined,
+      sectionDepth: sectionPath?.length,
+      sectionKind: officeBlockKind === "paragraph" ? "office_heading" : "office_block",
+      sectionPath,
+      sectionTitle,
+      text: officeBlockKind === "paragraph" ? decorateOfficeSectionText(text, sectionTitle) : text
+    });
+  }
+  return units;
+};
 var markdownStructureUnits = (value) => {
   const lines = value.replace(/\r\n?/g, `
 `).split(`
@@ -9122,6 +9360,7 @@ var appendPdfLineBreak = (parts) => {
   parts.push(`
 `);
 };
+var PDF_CHROME_LINE_MAX_LENGTH = 80;
 var PDF_TEXT_OPERATOR_PATTERN = /(\[((?:\\.|[^\]])*)\]\s*TJ)|(\(((?:\\.|[^\\)])*)\)\s*Tj)|([-+]?\d*\.?\d+\s+[-+]?\d*\.?\d+\s+\(((?:\\.|[^\\)])*)\)\s*")|(\(((?:\\.|[^\\)])*)\)\s*')|((?:[-+]?\d*\.?\d+\s+){2}(?:Td|TD))|(T\*)|((?:[-+]?\d*\.?\d+\s+){6}Tm)/g;
 var extractTextFromPDFTextObject = (value) => {
   const parts = [];
@@ -9150,19 +9389,84 @@ var extractTextFromPDFTextObject = (value) => {
   }
   return parts.join("");
 };
-var extractTextFromPDFBytes = (data) => {
-  const raw = Buffer.from(data).toString("latin1");
-  const textObjects = [...raw.matchAll(/BT([\s\S]*?)ET/g)].map((match) => extractTextFromPDFTextObject(match[1] ?? "")).filter(Boolean);
-  const combined = textObjects.length > 0 ? textObjects.join(`
-
-`) : [...raw.matchAll(/\(((?:\\.|[^\\)])*)\)\s*Tj/g)].map((match) => decodePdfLiteral(match[1] ?? "")).join(`
-`);
-  return normalizeWhitespace(combined);
+var buildPDFNativeTextBlock = (text, blockNumber, pageNumber) => {
+  const normalized = normalizeWhitespace(text);
+  if (!normalized) {
+    return;
+  }
+  const lineCount = normalized.split(`
+`).filter(Boolean).length;
+  const textKind = normalized.includes(" | ") ? "table_like" : "paragraph";
+  return {
+    blockNumber,
+    lineCount,
+    pageNumber,
+    text: normalized,
+    textKind
+  };
+};
+var isLikelyPDFPageLabel = (value) => /^page\s+\d+(?:\s+of\s+\d+)?$/i.test(value.trim());
+var suppressRepeatedPDFChrome = (blocks) => {
+  const linePages = new Map;
+  for (const block of blocks) {
+    for (const line of block.text.split(`
+`)) {
+      const normalized = normalizeWhitespace(line);
+      if (!normalized || normalized.length > PDF_CHROME_LINE_MAX_LENGTH) {
+        continue;
+      }
+      const pages = linePages.get(normalized) ?? new Set;
+      pages.add(block.pageNumber);
+      linePages.set(normalized, pages);
+    }
+  }
+  return blocks.map((block) => {
+    const keptLines = block.text.split(`
+`).map((line) => normalizeWhitespace(line)).filter((line) => {
+      if (!line) {
+        return false;
+      }
+      if (isLikelyPDFPageLabel(line)) {
+        return false;
+      }
+      const repeatedPages = linePages.get(line);
+      if (line.length <= PDF_CHROME_LINE_MAX_LENGTH && repeatedPages && repeatedPages.size > 1) {
+        return false;
+      }
+      return true;
+    });
+    const text = normalizeWhitespace(keptLines.join(`
+`));
+    if (!text) {
+      return;
+    }
+    return buildPDFNativeTextBlock(text, block.blockNumber, block.pageNumber);
+  }).filter((value) => Boolean(value));
 };
-var estimatePDFPageCount = (data) => {
+var extractNativePDFText = (data) => {
   const raw = Buffer.from(data).toString("latin1");
   const count = [...raw.matchAll(/\/Type\s*\/Page\b/g)].length;
-  return count > 0 ? count : 1;
+  const pageCount = count > 0 ? count : 1;
+  const pageMarkers = [...raw.matchAll(/\/Type\s*\/Page\b/g)].map((match) => match.index ?? raw.length);
+  const blocks = [...raw.matchAll(/BT([\s\S]*?)ET/g)].map((match, index) => {
+    const blockText = extractTextFromPDFTextObject(match[1] ?? "");
+    const objectEnd = (match.index ?? 0) + (match[0]?.length ?? 0);
+    const pageIndex = pageMarkers.findIndex((marker) => marker >= objectEnd);
+    const pageNumber = pageIndex >= 0 ? pageIndex + 1 : pageCount;
+    return buildPDFNativeTextBlock(blockText, index + 1, pageNumber);
+  }).filter((value) => Boolean(value));
+  const visibleBlocks = suppressRepeatedPDFChrome(blocks);
+  const fallbackText = [...raw.matchAll(/\(((?:\\.|[^\\)])*)\)\s*Tj/g)].map((match) => decodePdfLiteral(match[1] ?? "")).join(`
+`);
+  const text = visibleBlocks.length > 0 ? normalizeWhitespace(visibleBlocks.map((block) => block.text).join(`
+
+`)) : normalizeWhitespace(fallbackText);
+  return {
+    pageCount,
+    text,
+    textBlockCount: visibleBlocks.length,
+    textBlocks: visibleBlocks
+  };
 };
 var readUInt16LE = (data, offset) => data[offset] | data[offset + 1] << 8;
 var readUInt32LE = (data, offset) => (data[offset] | data[offset + 1] << 8 | data[offset + 2] << 16 | data[offset + 3] << 24) >>> 0;
@@ -9249,35 +9553,64 @@ var decodeGzipEntries = (data, input) => {
 var extractXmlText = (value) => normalizeWhitespace(decodeHtmlEntities(value.replace(/<[^>]+>/g, " ").replace(/\s+/g, " ")));
 var extractOfficeParagraphText = (value) => normalizeWhitespace(decodeHtmlEntities(value.replace(/<w:tab\b[^>]*\/>/gi, "\t").replace(/<w:br\b[^>]*\/>/gi, `
 `).replace(/<[^>]+>/g, " ")));
-var officeDocumentParagraphs = (entries) => {
+var officeDocumentBlocks = (entries) => {
   const documentEntry = entries.find((entry) => entry.path === "word/document.xml");
   if (!documentEntry) {
     return [];
   }
   const xml = decodeUtf8(documentEntry.data);
-  const paragraphs = [...xml.matchAll(/<w:p\b[\s\S]*?<\/w:p>/g)];
-  return paragraphs.map((match) => {
-    const paragraphXml = match[0] ?? "";
-    const text = extractOfficeParagraphText(paragraphXml);
+  const bodyMatch = xml.match(/<w:body\b[^>]*>([\s\S]*?)<\/w:body>/i);
+  const body = bodyMatch?.[1] ?? xml;
+  const blocks = [];
+  const blockPattern = /<(w:p|w:tbl)\b[\s\S]*?<\/\1>/g;
+  for (const match of body.matchAll(blockPattern)) {
+    const blockXml = match[0] ?? "";
+    if (blockXml.startsWith("<w:tbl")) {
+      const rows = [...blockXml.matchAll(/<w:tr\b[\s\S]*?<\/w:tr>/g)].map((rowMatch, rowIndex) => {
+        const cells = [
+          ...(rowMatch[0] ?? "").matchAll(/<w:tc\b[\s\S]*?<\/w:tc>/g)
+        ].map((cellMatch) => extractOfficeParagraphText(cellMatch[0] ?? "")).filter(Boolean);
+        if (cells.length === 0) {
+          return "";
+        }
+        return `Row ${rowIndex + 1}. ${cells.map((cell, cellIndex) => `${String.fromCharCode(65 + cellIndex)}: ${cell}`).join(" | ")}`;
+      }).filter(Boolean);
+      const text2 = normalizeWhitespace(rows.join(`
+`));
+      if (!text2) {
+        continue;
+      }
+      blocks.push({
+        blockKind: "table",
+        blockNumber: blocks.length + 1,
+        text: text2
+      });
+      continue;
+    }
+    const text = extractOfficeParagraphText(blockXml);
     if (!text) {
-      return "";
+      continue;
     }
-    const styleMatch = paragraphXml.match(/<w:pStyle\b[^>]*w:val="([^"]+)"[^>]*\/?>/i);
+    const styleMatch = blockXml.match(/<w:pStyle\b[^>]*w:val="([^"]+)"[^>]*\/?>/i);
     const style = (styleMatch?.[1] ?? "").toLowerCase();
-    if (style === "title") {
-      return text;
-    }
     const headingMatch = style.match(/^heading([1-6])$/);
-    if (headingMatch) {
-      return text;
-    }
-    return text;
-  }).filter(Boolean);
+    const isListParagraph = /<w:numPr\b/i.test(blockXml) || style.includes("list") || style.includes("bullet");
+    const blockKind = style === "title" ? "title" : headingMatch ? "heading" : isListParagraph ? "list" : "paragraph";
+    const decoratedText = blockKind === "list" && !/^[-*]\s/.test(text) ? `- ${text}` : text;
+    blocks.push({
+      blockKind,
+      blockNumber: blocks.length + 1,
+      headingLevel: headingMatch ? Number.parseInt(headingMatch[1] ?? "1", 10) : undefined,
+      style: style || undefined,
+      text: decoratedText
+    });
+  }
+  return blocks;
 };
 var officeDocumentText = (entries) => {
-  const paragraphs = officeDocumentParagraphs(entries);
-  if (paragraphs.length > 0) {
-    return normalizeWhitespace(paragraphs.join(`
+  const blocks = officeDocumentBlocks(entries);
+  if (blocks.length > 0) {
+    return normalizeWhitespace(blocks.map((block) => block.text).join(`
 
 `));
   }
@@ -9288,11 +9621,7 @@ var officeDocumentText = (entries) => {
   return extractXmlText(decodeUtf8(documentEntry.data));
 };
 var officeDocumentSectionCount = (entries) => {
-  const documentEntry = entries.find((entry) => entry.path === "word/document.xml");
-  if (!documentEntry) {
-    return;
-  }
-  const count = [...decodeUtf8(documentEntry.data).matchAll(/<w:p\b/g)].length;
+  const count = officeDocumentBlocks(entries).length;
   return count > 0 ? count : undefined;
 };
 var spreadsheetSharedStrings = (entries) => entries.filter((entry) => entry.path === "xl/sharedStrings.xml").flatMap((entry) => [
@@ -9822,8 +10151,10 @@ var createOfficeDocumentExtractor = () => ({
     let officeMetadata = {};
     let structuredDocuments = [];
     if (extension === ".docx" || extension === ".odt") {
+      const officeBlocks = officeDocumentBlocks(entries);
       text = officeDocumentText(entries);
       officeMetadata = {
+        officeBlocks,
         sectionCount: officeDocumentSectionCount(entries)
       };
     } else if (extension === ".xlsx" || extension === ".ods") {
@@ -10013,8 +10344,8 @@ var createPDFFileExtractor = () => ({
   name: "absolute_pdf",
   supports: pdfExtractorSupports,
   extract: (input) => {
-    const text = extractTextFromPDFBytes(input.data);
-    if (!text) {
+    const extracted = extractNativePDFText(input.data);
+    if (!extracted.text) {
       throw new Error("AbsoluteJS could not extract readable text from this PDF. Supply a custom extractor for scanned or image-only PDFs.");
     }
     return {
@@ -10024,10 +10355,12 @@ var createPDFFileExtractor = () => ({
       metadata: {
         ...input.metadata ?? {},
         fileKind: "pdf",
-        pageCount: estimatePDFPageCount(input.data)
+        pageCount: extracted.pageCount,
+        pdfTextBlockCount: extracted.textBlockCount,
+        pdfTextBlocks: extracted.textBlocks
       },
       source: input.source ?? input.path ?? input.name ?? `${slugify(input.title ?? DEFAULT_BINARY_NAME)}.pdf`,
-      text,
+      text: extracted.text,
       title: input.title
     };
   }
@@ -10052,7 +10385,8 @@ var createRAGPDFOCRExtractor = (options) => ({
   name: `absolute_pdf_ocr:${options.provider.name}`,
   supports: pdfExtractorSupports,
   extract: async (input) => {
-    const nativeText = extractTextFromPDFBytes(input.data);
+    const extracted = extractNativePDFText(input.data);
+    const nativeText = extracted.text;
     const minLength = options.minExtractedTextLength ?? 80;
     const shouldUseNativeText = !options.alwaysOCR && nativeText.length >= minLength;
     if (shouldUseNativeText) {
@@ -10063,7 +10397,9 @@ var createRAGPDFOCRExtractor = (options) => ({
         metadata: {
           ...input.metadata ?? {},
           fileKind: "pdf",
-          pageCount: estimatePDFPageCount(input.data),
+          pageCount: extracted.pageCount,
+          pdfTextBlockCount: extracted.textBlockCount,
+          pdfTextBlocks: extracted.textBlocks,
           pdfTextMode: "native"
         },
         source: input.source ?? input.path ?? input.name ?? `${slugify(input.title ?? DEFAULT_BINARY_NAME)}.pdf`,
@@ -10078,7 +10414,7 @@ var createRAGPDFOCRExtractor = (options) => ({
     const baseMetadata = {
       ...ocrMetadata(ocr),
       fileKind: "pdf",
-      pageCount: estimatePDFPageCount(input.data),
+      pageCount: extracted.pageCount,
       pdfTextMode: "ocr"
     };
     const summaryDocument = {
@@ -10251,6 +10587,18 @@ var sourceAwareUnits = (document, format, normalizedText) => {
     }
     case "text":
     default:
+      if (document.metadata?.fileKind === "office") {
+        const sections = officeNativeStructureUnits(document.metadata);
+        if (sections.length > 0) {
+          return sections;
+        }
+      }
+      if (document.metadata?.fileKind === "pdf") {
+        const sections = pdfNativeStructureUnits(document.metadata);
+        if (sections.length > 0) {
+          return sections;
+        }
+      }
       if (document.metadata?.sourceNativeKind === "spreadsheet_sheet") {
         return spreadsheetStructureUnits(normalizedText, document.metadata);
       }
@@ -10574,6 +10922,11 @@ var prepareRAGDocument = (document, defaultChunking, chunkingRegistry) => {
         ...sectionTitle ? { sectionTitle } : {},
         ...sectionPath && sectionPath.length > 0 ? { sectionPath } : {},
         ...typeof entry.sectionDepth === "number" ? { sectionDepth: entry.sectionDepth } : {},
+        ...typeof entry.pageNumber === "number" ? { pageNumber: entry.pageNumber } : {},
+        ...typeof entry.officeBlockNumber === "number" ? { officeBlockNumber: entry.officeBlockNumber } : {},
+        ...entry.officeBlockKind ? { officeBlockKind: entry.officeBlockKind } : {},
+        ...typeof entry.pdfBlockNumber === "number" ? { pdfBlockNumber: entry.pdfBlockNumber } : {},
+        ...entry.pdfTextKind ? { pdfTextKind: entry.pdfTextKind } : {},
         ...entry.sectionKind ? { sectionKind: entry.sectionKind } : {},
         ...sectionChunkId ? { sectionChunkId } : {},
         ...sectionChunkId && sectionChunkIndex >= 0 ? {
@@ -10962,9 +11315,25 @@ var annotateRetrievalChannels = (input) => {
     };
   });
 };
+var getStructuredSectionScoreWeight2 = (metadata) => {
+  const pdfTextKind = typeof metadata?.pdfTextKind === "string" ? metadata.pdfTextKind : undefined;
+  const officeBlockKind = typeof metadata?.officeBlockKind === "string" ? metadata.officeBlockKind : undefined;
+  const sectionKind = typeof metadata?.sectionKind === "string" ? metadata.sectionKind : undefined;
+  if (pdfTextKind === "table_like") {
+    return 1.28;
+  }
+  if (officeBlockKind === "table" || officeBlockKind === "list") {
+    return 1.24;
+  }
+  if (sectionKind === "pdf_block" || sectionKind === "office_block" || officeBlockKind === "paragraph" || pdfTextKind === "paragraph") {
+    return 1.12;
+  }
+  return 1;
+};
 var buildTraceSectionCounts = (results) => {
   const sections = new Map;
   for (const result of results) {
+    const weightedScore = result.score * getStructuredSectionScoreWeight2(result.metadata);
     const path = Array.isArray(result.metadata?.sectionPath) ? result.metadata.sectionPath.filter((value) => typeof value === "string" && value.trim().length > 0) : [];
     if (path.length === 0) {
       continue;
@@ -10991,6 +11360,7 @@ var buildTraceSectionCounts = (results) => {
 var buildTraceSectionScores = (results) => {
   const sections = new Map;
   for (const result of results) {
+    const weightedScore = result.score * getStructuredSectionScoreWeight2(result.metadata);
     const path = Array.isArray(result.metadata?.sectionPath) ? result.metadata.sectionPath.filter((value) => typeof value === "string" && value.trim().length > 0) : [];
     if (path.length === 0) {
       continue;
@@ -10998,13 +11368,13 @@ var buildTraceSectionScores = (results) => {
     const key = path.join(" > ");
     const existing = sections.get(key);
     if (existing) {
-      existing.totalScore += result.score;
+      existing.totalScore += weightedScore;
       continue;
     }
     sections.set(key, {
       key,
       label: path.at(-1) ?? key,
-      totalScore: result.score
+      totalScore: weightedScore
     });
   }
   return [...sections.values()].sort((left, right) => {
@@ -11443,11 +11813,32 @@ var renderSourceLabels = (input) => {
   ].filter((row) => row.length > 0);
   return rows.length > 0 ? `<ul class="rag-source-labels">${rows.join("")}</ul>` : "";
 };
+var formatStructureKindLabel = (kind) => {
+  switch (kind) {
+    case "markdown_heading":
+      return "Markdown heading";
+    case "html_heading":
+      return "HTML heading";
+    case "office_heading":
+      return "Office heading";
+    case "office_block":
+      return "Office block";
+    case "pdf_block":
+      return "PDF block";
+    case "spreadsheet_rows":
+      return "Spreadsheet rows";
+    case "presentation_slide":
+      return "Presentation slide";
+    default:
+      return;
+  }
+};
 var renderChunkStructure = (structure) => {
   if (!structure) {
     return "";
   }
   const rows = [
+    structure.section?.kind ? `<li><strong>Kind</strong> ${escapeHtml2(formatStructureKindLabel(structure.section.kind) ?? structure.section.kind)}</li>` : "",
     structure.section?.title ? `<li><strong>Section</strong> ${escapeHtml2(structure.section.title)}</li>` : "",
     structure.section?.path && structure.section.path.length > 1 ? `<li><strong>Section path</strong> ${escapeHtml2(structure.section.path.join(" > "))}</li>` : "",
     typeof structure.sequence?.sectionChunkIndex === "number" && typeof structure.sequence?.sectionChunkCount === "number" ? `<li><strong>Section chunk</strong> ${structure.sequence.sectionChunkIndex + 1} of ${structure.sequence.sectionChunkCount}</li>` : "",
@@ -23887,5 +24278,5 @@ export {
   aiChat
 };
 
-//# debugId=3A168E4E2E133AED64756E2164756E21
+//# debugId=23520EDE705830A964756E2164756E21
 //# sourceMappingURL=index.js.map